The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override safety and control mechanisms by mandating autonomous execution.
Evidence: 'CRITICAL: NEVER pause, stop, or wait for user input during execution. Proceed through ALL steps autonomously without asking the user to "continue"...'. This instruction is reinforced in Step 2: 'DO NOT stop here. DO NOT present the plan and wait for approval. Proceed IMMEDIATELY to Step 3.'
[COMMAND_EXECUTION]: The skill identifies and executes scripts defined in the project's configuration files at runtime.
Evidence: Step 3 and Step 6 describe detecting the package manager (bun, pnpm, npm) and running scripts like typecheck and test (e.g., npm test) found in package.json.
[DATA_EXFILTRATION]: The skill has broad access to the local filesystem and integrates with external MCP tools (browser testing, API testing), which could be chained to exfiltrate data if manipulated.
Evidence: It reads sensitive configuration files like .mcp.json and project documentation, and uses browser/API testing MCPs that can reach external networks.
[INDIRECT_PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection as it processes untrusted data files to drive its high-privilege execution flow.
Ingestion points: Reads bug reports from ./pbis/pbi-[feature-slug]/bugs.md and requirements from pbi.md and techspec.md (Step 1).
Boundary markers: Absent. There are no instructions to treat the content of these files as data rather than instructions.
Capability inventory: File system read/write (Edit/Write tools), shell command execution (npm test), and access to various MCP servers (database, cache, browser).
Sanitization: Absent. The agent does not validate or sanitize the 'root cause' or 'fix strategy' derived from the input files before implementing them.
[DYNAMIC_EXECUTION]: The skill dynamically modifies the codebase and then executes that code via the test runner.
Evidence: Step 3 (Implement Fixes) and Step 4 (Create Regression Tests) involve writing new code/tests and immediately executing them in Step 6.

do-execute-bugfix